Limits...
Detecting Lung Diseases from Exhaled Aerosols: Non-Invasive Lung Diagnosis Using Fractal Analysis and SVM Classification.

Xi J, Zhao W, Yuan JE, Kim J, Si X, Xu X - PLoS ONE (2015)

Bottom Line: By employing the 10-fold cross-validation method, we achieved 100% classification accuracy among four asthmatic models using an ideal 108-sample dataset and 99.1% accuracy using a more realistic 324-sample dataset.The fractal-SVM classifier has been shown to be robust, highly sensitive to structural variations, and inherently suitable for investigating aerosol-disease correlations.For the first time, this study quantitatively linked the exhaled aerosol patterns with their underlying diseases and set the stage for the development of a computer-aided diagnostic system for non-invasive detection of obstructive respiratory diseases.

View Article: PubMed Central - PubMed

Affiliation: School of Engineering and Technology, Central Michigan University, Mount Pleasant, Michigan, United States of America.

ABSTRACT

Background: Each lung structure exhales a unique pattern of aerosols, which can be used to detect and monitor lung diseases non-invasively. The challenges are accurately interpreting the exhaled aerosol fingerprints and quantitatively correlating them to the lung diseases.

Objective and methods: In this study, we presented a paradigm of an exhaled aerosol test that addresses the above two challenges and is promising to detect the site and severity of lung diseases. This paradigm consists of two steps: image feature extraction using sub-regional fractal analysis and data classification using a support vector machine (SVM). Numerical experiments were conducted to evaluate the feasibility of the breath test in four asthmatic lung models. A high-fidelity image-CFD approach was employed to compute the exhaled aerosol patterns under different disease conditions.

Findings: By employing the 10-fold cross-validation method, we achieved 100% classification accuracy among four asthmatic models using an ideal 108-sample dataset and 99.1% accuracy using a more realistic 324-sample dataset. The fractal-SVM classifier has been shown to be robust, highly sensitive to structural variations, and inherently suitable for investigating aerosol-disease correlations.

Conclusion: For the first time, this study quantitatively linked the exhaled aerosol patterns with their underlying diseases and set the stage for the development of a computer-aided diagnostic system for non-invasive detection of obstructive respiratory diseases.

No MeSH data available.


Related in: MedlinePlus

Analysis of misclassified samples in the 324-sample-experiment which has a respiration range of 30±10 L/min and two upper airway variations.The prediction accuracy is 99.1% with three misclassified samples in total (a). The image, sample number, and test parameters of the misclassified samples are listed in (b). The locations of the three misclassified samples are marked in the distance matrix (c), and their zoom-in plots are shown in (d).
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4589383&req=5

pone.0139511.g008: Analysis of misclassified samples in the 324-sample-experiment which has a respiration range of 30±10 L/min and two upper airway variations.The prediction accuracy is 99.1% with three misclassified samples in total (a). The image, sample number, and test parameters of the misclassified samples are listed in (b). The locations of the three misclassified samples are marked in the distance matrix (c), and their zoom-in plots are shown in (d).

Mentions: We detected three misclassified samples in the 10-fold cross-validation with 324 samples (Fig 8A): S247 (a D3-feature sample being predicted as D2: D3 → D2), S97 (D2 → D0), S229 (D1 → D3). The images and test parameters are shown in Fig 8B. In order to discover the causes of these misclassifications, a detailed analysis was conducted on the above three samples. Fig 8C shows the locations of these three samples in the distance matrix and Fig 8D shows their zoom-in plots. Considering case I (S247), there were three outliers in the lower left corner, one of them being S247 (the third from the right). The other two were S134 and S204, both of which had a D2-feature. Due to large dissimilarities from all other samples, these three samples were clustered into one subgroup, which led to the misclassification of the S247 (D3) as D2. In light of case II (S97), the misclassification of the D2-feature sample as D0 was explained by the dendrogram in Fig 8D. This sample bordered between the D0 and D2 samples, but showed more affinity with the D0 sub-group.


Detecting Lung Diseases from Exhaled Aerosols: Non-Invasive Lung Diagnosis Using Fractal Analysis and SVM Classification.

Xi J, Zhao W, Yuan JE, Kim J, Si X, Xu X - PLoS ONE (2015)

Analysis of misclassified samples in the 324-sample-experiment which has a respiration range of 30±10 L/min and two upper airway variations.The prediction accuracy is 99.1% with three misclassified samples in total (a). The image, sample number, and test parameters of the misclassified samples are listed in (b). The locations of the three misclassified samples are marked in the distance matrix (c), and their zoom-in plots are shown in (d).
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4589383&req=5

pone.0139511.g008: Analysis of misclassified samples in the 324-sample-experiment which has a respiration range of 30±10 L/min and two upper airway variations.The prediction accuracy is 99.1% with three misclassified samples in total (a). The image, sample number, and test parameters of the misclassified samples are listed in (b). The locations of the three misclassified samples are marked in the distance matrix (c), and their zoom-in plots are shown in (d).
Mentions: We detected three misclassified samples in the 10-fold cross-validation with 324 samples (Fig 8A): S247 (a D3-feature sample being predicted as D2: D3 → D2), S97 (D2 → D0), S229 (D1 → D3). The images and test parameters are shown in Fig 8B. In order to discover the causes of these misclassifications, a detailed analysis was conducted on the above three samples. Fig 8C shows the locations of these three samples in the distance matrix and Fig 8D shows their zoom-in plots. Considering case I (S247), there were three outliers in the lower left corner, one of them being S247 (the third from the right). The other two were S134 and S204, both of which had a D2-feature. Due to large dissimilarities from all other samples, these three samples were clustered into one subgroup, which led to the misclassification of the S247 (D3) as D2. In light of case II (S97), the misclassification of the D2-feature sample as D0 was explained by the dendrogram in Fig 8D. This sample bordered between the D0 and D2 samples, but showed more affinity with the D0 sub-group.

Bottom Line: By employing the 10-fold cross-validation method, we achieved 100% classification accuracy among four asthmatic models using an ideal 108-sample dataset and 99.1% accuracy using a more realistic 324-sample dataset.The fractal-SVM classifier has been shown to be robust, highly sensitive to structural variations, and inherently suitable for investigating aerosol-disease correlations.For the first time, this study quantitatively linked the exhaled aerosol patterns with their underlying diseases and set the stage for the development of a computer-aided diagnostic system for non-invasive detection of obstructive respiratory diseases.

View Article: PubMed Central - PubMed

Affiliation: School of Engineering and Technology, Central Michigan University, Mount Pleasant, Michigan, United States of America.

ABSTRACT

Background: Each lung structure exhales a unique pattern of aerosols, which can be used to detect and monitor lung diseases non-invasively. The challenges are accurately interpreting the exhaled aerosol fingerprints and quantitatively correlating them to the lung diseases.

Objective and methods: In this study, we presented a paradigm of an exhaled aerosol test that addresses the above two challenges and is promising to detect the site and severity of lung diseases. This paradigm consists of two steps: image feature extraction using sub-regional fractal analysis and data classification using a support vector machine (SVM). Numerical experiments were conducted to evaluate the feasibility of the breath test in four asthmatic lung models. A high-fidelity image-CFD approach was employed to compute the exhaled aerosol patterns under different disease conditions.

Findings: By employing the 10-fold cross-validation method, we achieved 100% classification accuracy among four asthmatic models using an ideal 108-sample dataset and 99.1% accuracy using a more realistic 324-sample dataset. The fractal-SVM classifier has been shown to be robust, highly sensitive to structural variations, and inherently suitable for investigating aerosol-disease correlations.

Conclusion: For the first time, this study quantitatively linked the exhaled aerosol patterns with their underlying diseases and set the stage for the development of a computer-aided diagnostic system for non-invasive detection of obstructive respiratory diseases.

No MeSH data available.


Related in: MedlinePlus